earakely-scale/DBBrowserForSQLiteFromRequestComputerEnv-REA

DB Browser for SQLite: a computer-use environment built from a one-line request

The real DB Browser for SQLite 3.12 database editor running on an Ubuntu 24.04 desktop, as an agentenv computer environment. An agent completes the task only through screenshots, mouse and keyboard, and a verifier grades the saved files with no model.

Claude Code built this environment from a one-line request in about 24 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. DB Browser for SQLite opens a small contacts database, Documents/contacts.db, at launch: one table, customers, of six synthetic customers with their email, phone and city. Priya Raman's email is misspelled as priya.raman@exmaple.com. The agent has to correct it to priya.raman@example.com and write the change to the file. It sees only the screen and acts only with the mouse and keyboard.

Watch

• Agent runs (Claude Code with Claude Sonnet 5.5): run 1, run 2, run 3
• Witness, which proves the task can be done through the computer: witness.mp4

Each video is the computer's own screen recording at 10 frames per second.

The task

In DB Browser for SQLite, the customer contacts database Documents/contacts.db has the wrong email address for Priya Raman. Correct her email address to priya.raman@example.com and save the change.

bundle/artifacts/fix-customer-email-check/verify.py reads the app's files through the computer's data/get view and passes only if every check holds:

Check	Kind	File
check-1	row_present	Documents/contacts.db
rest-of-row	row_present	Documents/contacts.db
other-customer	row_present	Documents/contacts.db
no-rows-added	row_count	Documents/contacts.db

REA reads an SQLite database as tables of rows. The checks find rows by their id: customer 3 must have the email priya.raman@example.com and keep her name, phone and city, customer 5, Grace Whitfield, must keep her name and email, and the table must still hold six rows. REA's draft suggested the email check (check-1) as written; Claude Code added the other three. DB Browser keeps edits in a pending transaction until Write Changes (Ctrl+S), so an edit that is never written leaves the file as it was. The verifier grades once the database file has held still for 3 seconds.

Results

3 of 3 runs passed, about $0.51 per run.

Run	Outcome	Turns	Cost	Video
1	passed	12	$0.76	video
2	passed	13	$0.26	video
3	passed	12	$0.51	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (sqlitebrowser 3.12.2).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/DBBrowserForSQLiteFromRequestComputerEnv-REA --repo-type dataset --local-dir dbbrowser-request
agent-env run dbbrowser-request/bundle --task fix-customer-email --agent claude-code --model <model>

The agent needs a model provider configured for agentenv. --task smoke only boots the computer, with no agent. bundle/README.md describes the bundle in detail.

How it was made

Claude Code built this environment from a one-line request, given only REA's build skill and REA's MCP tools:

Build me a DB Browser for SQLite environment where the agent corrects a customer's email address in a small contacts database and saves it.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 24 minutes.

It took a second capture. On the first, Claude Code named no state paths, since DB Browser's window title already showed the database. REA then kept no starting copy of the database, which DB Browser held open, so it could not read the file the same way at both ends and drafted no checks. Claude Code captured again with the database's folder named as a state path, and REA's draft suggested the email check. The steps were REA's usual ones:

1. Capture. REA launched DB Browser for SQLite on a Linux desktop and recorded its package, windows and opened files while the edit was made. It also recorded the app's state folders before and after.
2. Export. REA turned the capture into an agentenv computer_app. It installs the app, seeds the starting files, starts the app, and waits for its main window. It also serves typed views of the app's files for verifiers.
3. Task. The task's checks were validated against the capture: they fail on the starting state and pass on the captured end state.
4. Witness. In a fresh computer, the edit was made through computer actions alone, and the checks were read at four stages. They held on the first try:

Stage	Checks
seed	fail, as expected: Priya Raman's email is priya.raman@exmaple.com; the other checks pass
after_edit	pass
near_miss	fail, as expected: the email was typed as priya.raman@example.co
after_reset	fail, as expected: data/reset restored the seed

The agent runs came after the build: Claude Code with Claude Sonnet 5.5 (--model anthropic/claude-sonnet-5-5 --no-emulate, on an amd64 host) solved the task 3 times, each run recorded.

What's inside

bundle/      the agentenv bundle: envs/ (app module, seed files), tasks/, artifacts/ (verifier), evals/
media/       witness and agent-run recordings (MP4) with short GIF previews
results.json the runs above, with each check's result

Limits

• Only Priya Raman's row and Grace Whitfield's name and email are checked, so an edit to one of the other four customers still passes.
• The computer installs whatever sqlitebrowser Ubuntu 24.04's archive holds at build time (3.12.2 for these runs); it is not pinned.
• Three runs is a small sample: the solve rate shows the task is solvable, not how hard it is.

Licences

The bundle code, the task and the synthetic seed files are Apache-2.0. DB Browser for SQLite itself is not included: the computer installs it from Ubuntu's archive (sqlitebrowser 3.12.2) at build time, under its own licence (MPL-2.0 or GPL-3.0-or-later).